Read the literature, e.g. http://hal.pratt.duke.edu/publications

There's a huge difference between mostly working (especially on a highway) and truly autonomous door-to-door.

Sorry, which literature in particular? That page has a variety of posted papers, most of which cover things unrelated to practical autonomous road vehicles.

You also don't seem to have addressed my point that there are cars on the road right now with no driver behind the wheel. We are long past the point of mostly working and very close to the point of autonomous door-to-door in favorable cities.

In one city, with really good road paint, street signs, and incredibly detailed maps compared to everywhere else, almost no rains, and never any snow. And those cars are rare. Try making 10% of the cars on the road made up of them, all of which go to the extra safe fallback mode at a particular intersection; 1 out of 10 cars doing that will annoy the hell out of 8 of the other human driven cars. Now try 30% of the cars made up of cars in fallback.

I think autonomy will be partial, conditional, and thus not door to door autonomous for at least a decade.

I sooner expect Alphabet to sell this tech to a big 10 auto manufacturer, or to Apple, for the same reasons as Boston Dynamics: where's the cashflow? If it's not up for sale in 5 years, I'll expect it's the real deal and we'll see door to door in another 10 years.

Yes, but the point is it doesn't need to be ready for 100% of locations immediately. Los Angeles alone has close to 19 million people. Add in other favorable cities like San Jose/SV, San Diego, Las Vegas, and Phoenix and you're looking at a fair portion of the total US population. You can have robotaxis serving those cities initially while the technology matures, and location-restrict autonomous driving for consumer-owned cars to the same places. People seem to think that the moment autonomous cars hit the road, they need to be able to handle a blizzard on a dirt road in upstate New York, but that just doesn't make any sense.